Cheng Chi

Cheng Chi, Amine Mohamed Aboussalah, Elias B. Khalil et al. · utoronto

Column Generation (CG) is an iterative algorithm for solving linear programs (LPs) with an extremely large number of variables (columns). CG is the workhorse for tackling large-scale \textit{integer} linear programs, which rely on CG to solve LP relaxations within a branch and price algorithm. Two canonical applications are the Cutting Stock Problem (CSP) and Vehicle Routing Problem with Time Windows (VRPTW). In VRPTW, for example, each binary variable represents the decision to include or exclude a \textit{route}, of which there are exponentially many; CG incrementally grows the subset of columns being used, ultimately converging to an optimal solution. We propose RLCG, the first Reinforcement Learning (RL) approach for CG. Unlike typical column selection rules which myopically select a column based on local information at each iteration, we treat CG as a sequential decision-making problem: the column selected in a given iteration affects subsequent column selections. This perspective lends itself to a Deep Reinforcement Learning approach that uses Graph Neural Networks (GNNs) to represent the variable-constraint structure in the LP of interest. We perform an extensive set of experiments using the publicly available BPPLIB benchmark for CSP and Solomon benchmark for VRPTW. RLCG converges faster and reduces the number of CG iterations by 22.4\% for CSP and 40.9\% for VRPTW on average compared to a commonly used greedy policy. Our code is available at https://github.com/chichengmessi/reinforcement-learning-for-column-generation.git.

Mengda Xu, Zhenjia Xu, Cheng Chi et al.

Human demonstration videos are a widely available data source for robot learning and an intuitive user interface for expressing desired behavior. However, directly extracting reusable robot manipulation skills from unstructured human videos is challenging due to the big embodiment difference and unobserved action parameters. To bridge this embodiment gap, this paper introduces XSkill, an imitation learning framework that 1) discovers a cross-embodiment representation called skill prototypes purely from unlabeled human and robot manipulation videos, 2) transfers the skill representation to robot actions using conditional diffusion policy, and finally, 3) composes the learned skill to accomplish unseen tasks specified by a human prompt video. Our experiments in simulation and real-world environments show that the discovered skill prototypes facilitate both skill transfer and composition for unseen tasks, resulting in a more general and scalable imitation learning framework. The benchmark, code, and qualitative results are on https://xskill.cs.columbia.edu/

Shihan Wu, Xuecheng Liu, Shaoxuan Xie et al.

Despite the critical role of bimanual manipulation in endowing robots with human-like dexterity, large-scale and diverse datasets remain scarce due to the significant hardware heterogeneity across bimanual robotic platforms. To bridge this gap, we introduce RoboCOIN, a large-scale multi-embodiment bimanual manipulation dataset comprising over 180,000 demonstrations collected from 15 distinct robotic platforms. Spanning 16 diverse environments-including residential, commercial, and industrial settings-the dataset features 421 bimanual tasks systematically categorized by 39 bimanual collaboration actions and 432 objects. A key innovation of our work is the hierarchical capability pyramid, which provides granular annotations ranging from trajectory-level concepts to segment-level subtasks and frame-level kinematics. Furthermore, we present CoRobot, an efficient data processing pipeline powered by the Robot Trajectory Markup Language (RTML), designed to facilitate quality assessment, automated annotation, and unified multi-embodiment and data management. Extensive experiments demonstrate the effectiveness of RoboCOIN in enhancing the performance of various bimanual manipulation models across a wide spectrum of robotic embodiments. The entire dataset and codebase are fully open-sourced, providing a valuable resource for advancing research in bimanual and multi-embodiment manipulation.

Shuanghao Bai, Jing Lyu, Wanqi Zhou et al.

Vision-Language-Action (VLA) models benefit from chain-of-thought (CoT) reasoning, but existing approaches incur high inference overhead and rely on discrete reasoning representations that mismatch continuous perception and control. We propose Latent Reasoning VLA (LaRA-VLA), a unified VLA framework that internalizes multi-modal CoT reasoning into continuous latent representations for embodied action. LaRA-VLA performs unified reasoning and prediction in latent space, eliminating explicit CoT generation at inference time and enabling efficient, action-oriented control. To realize latent embodied reasoning, we introduce a curriculum-based training paradigm that progressively transitions from explicit textual and visual CoT supervision to latent reasoning, and finally adapts latent reasoning dynamics to condition action generation. We construct two structured CoT datasets and evaluate LaRA-VLA on both simulation benchmarks and long-horizon real-robot manipulation tasks. Experimental results show that LaRA-VLA consistently outperforms state-of-the-art VLA methods while reducing inference latency by up to 90\% compared to explicit CoT-based approaches, demonstrating latent reasoning as an effective and efficient paradigm for real-time embodied control. Project Page: https://loveju1y.github.io/Latent-Reasoning-VLA/

Mengda Xu, Zhenjia Xu, Yinghao Xu et al.

We present Im2Flow2Act, a scalable learning framework that enables robots to acquire real-world manipulation skills without the need of real-world robot training data. The key idea behind Im2Flow2Act is to use object flow as the manipulation interface, bridging domain gaps between different embodiments (i.e., human and robot) and training environments (i.e., real-world and simulated). Im2Flow2Act comprises two components: a flow generation network and a flow-conditioned policy. The flow generation network, trained on human demonstration videos, generates object flow from the initial scene image, conditioned on the task description. The flow-conditioned policy, trained on simulated robot play data, maps the generated object flow to robot actions to realize the desired object movements. By using flow as input, this policy can be directly deployed in the real world with a minimal sim-to-real gap. By leveraging real-world human videos and simulated robot play data, we bypass the challenges of teleoperating physical robots in the real world, resulting in a scalable system for diverse tasks. We demonstrate Im2Flow2Act's capabilities in a variety of real-world tasks, including the manipulation of rigid, articulated, and deformable objects.

Yipu Wang, Yuheng Ji, Yuyang Liu et al.

Cross-view correspondence is a fundamental capability for spatial understanding and embodied AI. However, it is still far from being realized in Vision-Language Models (VLMs), especially in achieving precise point-level correspondence, which is crucial for precise affordance interaction. So we propose the Cross-View Point Correspondence (CVPC) task and CrossPoint-Bench, a comprehensive benchmark with hierarchical design, inspired by the human cognitive process of "perceive", "reason", and "correspond". Our evaluation shows the state-of-the-art models (e.g., Gemini-2.5-Pro) still fall far behind humans, with a gap of over 54.65% in overall accuracy, exposing a challenge in transitioning from coarse-grained judgement to fine-grained coordinate prediction. To address this problem, we construct CrossPoint-378K, a dataset with 378K question-answering pairs across 900 scenes, focused on actionable affordance regions that better reflect real-world manipulation and interaction scenarios. Furthermore, we propose CroPond that trained on the CrossPoint-378K dataset. Our CroPond achieves state-of-the-art performance on CrossPoint-Bench, surpassing Gemini-2.5-Pro by 39.7% accuracy, which offers a foundation for advancing future work on cross-view correspondence. The benchmark, dataset, and model are publicly available at https://github.com/WangYipu2002/CrossPoint.

Yun Wang, Cheng Chi, Xin Yang

Scene flow estimation, which predicts the 3D motion of scene points from point clouds, is a core task in autonomous driving and many other 3D vision applications. Existing methods either suffer from structure distortion due to ignorance of rigid motion consistency or require explicit pose estimation and 3D object segmentation. Errors of estimated poses and segmented objects would yield inaccurate rigidity constraints and in turn mislead scene flow estimation. In this paper, we propose a novel weight-sharing aggregation (WSA) method for feature and scene flow up-sampling. WSA does not rely on estimated poses and segmented objects, and can implicitly enforce rigidity constraints to avoid structure distortion in scene flow estimation. To further exploit geometric information and preserve local structure, we design a deformation degree module aim to keep the local region invariance. We modify the PointPWC-Net and integrate the proposed WSA and deformation degree module into the enhanced PointPWC-Net to derive an end-to-end scene flow estimation network, called WSAFlowNet. Extensive experimental results on the FlyingThings3D and KITTI datasets demonstrate that our WSAFlowNet achieves the state-of-the-art performance and outperforms previous methods by a large margin. We will release the source code at https://github.com/wangyunlhr/WSAFlowNet.git.

Amine Mohamed Aboussalah, Min-Jae Kwon, Raj G Patel et al.

Time series observations can be seen as realizations of an underlying dynamical system governed by rules that we typically do not know. For time series learning tasks, we need to understand that we fit our model on available data, which is a unique realized history. Training on a single realization often induces severe overfitting lacking generalization. To address this issue, we introduce a general recursive framework for time series augmentation, which we call Recursive Interpolation Method, denoted as RIM. New samples are generated using a recursive interpolation function of all previous values in such a way that the enhanced samples preserve the original inherent time series dynamics. We perform theoretical analysis to characterize the proposed RIM and to guarantee its test performance. We apply RIM to diverse real world time series cases to achieve strong performance over non-augmented data on regression, classification, and reinforcement learning tasks.

Zirui Ge, Pengxiang Ding, Baohua Yin et al.

Video action models are an appealing foundation for Vision--Language--Action systems because they can learn visual dynamics from large-scale video data and transfer this knowledge to downstream robot control. Yet current diffusion-based video predictors are trained with likelihood-surrogate objectives, which encourage globally plausible predictions without explicitly optimizing the precision-critical visual dynamics needed for manipulation. This objective mismatch often leads to subtle errors in object pose, spatial relations, and contact timing that can be amplified by downstream policies. We propose VAMPO, a post-training framework that directly improves visual dynamics in video action models through policy optimization. Our key idea is to formulate multi-step denoising as a sequential decision process and optimize the denoising policy with rewards defined over expert visual dynamics in latent space. To make this optimization practical, we introduce an Euler Hybrid sampler that injects stochasticity only at the first denoising step, enabling tractable low-variance policy-gradient estimation while preserving the coherence of the remaining denoising trajectory. We further combine this design with GRPO and a verifiable non-adversarial reward. Across diverse simulated and real-world manipulation tasks, VAMPO improves task-relevant visual dynamics, leading to better downstream action generation and stronger generalization. The homepage is https://vampo-robot.github.io/VAMPO/.

Yuheng Ji, Yuyang Liu, Huajie Tan et al.

Current robotic evaluation is still largely dominated by binary success rates, which collapse rich execution processes into a single outcome and obscure critical qualities such as progress, efficiency, and stability. To address this limitation, we propose PRM-as-a-Judge, a dense evaluation paradigm that leverages Process Reward Models (PRMs) to audit policy execution directly from trajectory videos by estimating task progress from observation sequences. Central to this paradigm is the OPD (Outcome-Process-Diagnosis) metric system, which explicitly formalizes execution quality via a task-aligned progress potential. We characterize dense robotic evaluation through two axiomatic properties: macro-consistency, which requires additive and path-consistent aggregation, and micro-resolution, which requires sensitivity to fine-grained physical evolution. Under this formulation, potential-based PRM judges provide a natural instantiation of dense evaluation, with macro-consistency following directly from the induced scalar potential. We empirically validate the micro-resolution property using RoboPulse, a diagnostic benchmark specifically designed for probing micro-scale progress discrimination, where several trajectory-trained PRM judges outperform discriminative similarity-based methods and general-purpose foundation-model judges. Finally, leveraging PRM-as-a-Judge and the OPD metric system, we conduct a structured audit of mainstream policy paradigms across long-horizon tasks, revealing behavioral signatures and failure modes that are invisible to outcome-only metrics.

Hongjing Wu, Cheng Chi, Jinlin Wu et al.

Camera-only 3D object detection is critical for autonomous driving, offering a cost-effective alternative to LiDAR based methods. In particular, multi-view 3D object detection has emerged as a promising direction due to its balanced trade-off between performance and cost. However, existing methods often suffer significant performance degradation under complex environmental conditions such as nighttime, fog, and rain, primarily due to their reliance on training data collected mostly in ideal conditions. To address this challenge, we propose UniDA3D, a unified domain-adaptive multi-view 3D object detector designed for robust perception under diverse adverse conditions. UniDA3D formulates nighttime, rainy, and foggy scenes as a unified multi target domain adaptation problem and leverages a novel query guided domain discrepancy mitigation (QDDM) module to align object features between source and target domains at both batch and global levels via query-centric adversarial and contrastive learning. Furthermore, we introduce a domain-adaptive teacher student training pipeline with an exponential-moving-average teacher and dynamically updated high-quality pseudo labels to enhance consistency learning and suppress background noise in unlabeled target domains. In contrast to prior approaches that require separate training for each condition, UniDA3D performs a single unified training process across multiple domains, enabling robust all-weather 3D perception. On a synthesized multi-view 3D benchmark constructed by generating nighttime, rainy, and foggy counterparts from nuScenes (nuScenes-Night, nuScenes-Rain, and nuScenes-Haze), UniDA3D consistently outperforms state of-the-art camera-only multi-view 3D detectors under extreme conditions, achieving substantial gains in mAP and NDS while maintaining real-time inference efficiency.

Lijun Zhou, Hongcheng Luo, Zhenxin Zhu et al.

This report presents a unified technical system addressing the two core capabilities of world models for autonomous driving: world representation and world generation. For world representation, we propose WorldRec, a feed-forward reconstruction architecture driven by sparse scene queries. WorldRec initializes structured queries in 3D space, leveraging them to aggregate cross-view, cross-temporal features, thereby naturally enforcing spatial consistency across frames and yielding compact yet high-fidelity 3D Gaussian scene representations. For world generation, we propose WorldGen, a two-stage training framework of bidirectional pretraining followed by causal fine-tuning through three progressive stages (Teacher Forcing, ODE distillation, and DMD), enabling high-quality online causal video generation in as few as 4 denoising steps. Building on both modules, we further introduce the JWM, which deeply integrates WorldRec and WorldGen to achieve synergistic gains in generation stability, cross-frame consistency, and visual fidelity, providing a solid foundation for closed-loop simulation, data synthesis, and end-to-end training in autonomous driving.

Mengzhen Liu, Enshen Zhou, Cheng Chi et al.

Active perception and manipulation are crucial for robots to interact with complex scenes. Existing methods struggle to unify semantic-driven active perception with robust, viewpoint-invariant execution. We propose SaPaVe, an end-to-end framework that jointly learns these capabilities in a data-efficient manner. Our approach decouples camera and manipulation actions rather than placing them in a shared action space, and follows a bottom-up training strategy: we first train semantic camera control on a large-scale dataset, then jointly optimize both action types using hybrid data. To support this framework, we introduce ActiveViewPose-200K, a dataset of 200k image-language-camera movement pairs for semantic camera movement learning, and a 3D geometry-aware module that improves execution robustness under dynamic viewpoints. We also present ActiveManip-Bench, the first benchmark for evaluating active manipulation beyond fixed-view settings. Extensive experiments in both simulation and real-world environments show that SaPaVe outperforms recent vision-language-action models such as GR00T N1 and \(π_0\), achieving up to 31.25\% higher success rates in real-world tasks. These results show that tightly coupled perception and execution, when trained with decoupled yet coordinated strategies, enable efficient and generalizable active manipulation. Project page: https://lmzpai.github.io/SaPaVe

Zhe Li, Cheng Chi, Boan Zhu et al.

Humans learn locomotion through visual observation, interpreting visual content first before imitating actions. However, state-of-the-art humanoid locomotion systems rely on either curated motion capture trajectories or sparse text commands, leaving a critical gap between visual understanding and control. Text-to-motion methods suffer from semantic sparsity and staged pipeline errors, while video-based approaches only perform mechanical pose mimicry without genuine visual understanding. We propose RoboMirror, the first retargeting-free video-to-locomotion framework embodying "understand before you imitate". Leveraging VLMs, it distills raw egocentric/third-person videos into visual motion intents, which directly condition a diffusion-based policy to generate physically plausible, semantically aligned locomotion without explicit pose reconstruction or retargeting. Extensive experiments validate the effectiveness of RoboMirror, it enables telepresence via egocentric videos, drastically reduces third-person control latency by 80%, and achieves a 3.7% higher task success rate than baselines. By reframing humanoid control around video understanding, we bridge the visual understanding and action gap.

Yuheng Ji, Huajie Tan, Cheng Chi et al.

We introduce \textsc{MathSticks}, a benchmark for Visual Symbolic Compositional Reasoning (VSCR), which unifies visual perception, symbolic manipulation, and arithmetic consistency. Each task presents an incorrect matchstick equation that must be corrected by moving one or two sticks under strict conservation rules. The benchmark includes both text-guided and purely visual settings, systematically covering digit scale, move complexity, solution multiplicity, and operator variation, with 1.4M generated instances and a curated test set. Evaluations of 14 vision--language models reveal substantial limitations: closed-source models succeed only on simple cases, open-source models fail in the visual regime, while humans exceed 90\% accuracy. These findings establish \textsc{MathSticks} as a rigorous testbed for advancing compositional reasoning across vision and symbols. Our code and dataset are publicly available at https://github.com/Yuheng2000/MathSticks.

Cheng Chi, Xianqi Wang, Hongcheng Luo et al.

High-fidelity reconstruction of driving scenes is crucial for autonomous driving. While recent feedforward 3D Gaussian Splatting (3DGS) methods enable fast reconstruction, their per-pixel Gaussian prediction paradigm often suffers from multi-view inconsistency and layering artifacts. Moreover, existing methods often model dynamic instances via dense flow prediction, which lacks explicit cross-view correspondence and instance-level consistency. In this paper, we propose PointForward, a feedforward driving reconstruction framework through point-aligned representations. Unlike pixel-aligned methods, we initialize sparse 3D queries in world space and aggregate multi-view image information via spatial-temporal fusion onto these queries, enforcing explicit cross-view consistency in a single feedforward pass. To handle scene dynamics, we introduce scene graphs that explicitly organize moving instances during reconstruction. By leveraging 3D bounding boxes, our method enables instance-level motion propagation and temporally consistent dynamic representations. Extensive experiments demonstrate that PointForward achieves state-of-the-art performance on large-scale driving benchmarks. The code will be available upon the publication of the paper.

Kry Yik Chau Lui, Cheng Chi, Kishore Basu et al.

Despite their dominance in vision and language, deep neural networks often underperform relative to tree-based models on tabular data. To bridge this gap, we incorporate five key inductive biases into deep learning: robustness to irrelevant features, axis alignment, localized irregularities, feature heterogeneity, and training stability. We propose \emph{LassoFlexNet}, an architecture that evaluates the linear and nonlinear marginal contribution of each input via Per-Feature Embeddings, and sparsely selects relevant variables using a Tied Group Lasso mechanism. Because these components introduce optimization challenges that destabilize standard proximal methods, we develop a \emph{Sequential Hierarchical Proximal Adaptive Gradient optimizer with exponential moving averages (EMA)} to ensure stable convergence. Across $52$ datasets from three benchmarks, LassoFlexNet matches or outperforms leading tree-based models, achieving up to a $10$\% relative gain, while maintaining Lasso-like interpretability. We substantiate these empirical results with ablation studies and theoretical proofs confirming the architecture's enhanced expressivity and structural breaking of undesired rotational invariance.

Cheng Chi

Controlling continuous-time dynamical systems is generally a two step process: first, identify or model the system dynamics with differential equations, then, minimize the control objectives to achieve optimal control function and optimal state trajectories. However, any inaccuracy in dynamics modeling will lead to sub-optimality in the resulting control function. To address this, we propose a neural ODE based method for controlling unknown dynamical systems, denoted as Neural Control (NC), which combines dynamics identification and optimal control learning using a coupled neural ODE. Through an intriguing interplay between the two neural networks in coupled neural ODE structure, our model concurrently learns system dynamics as well as optimal controls that guides towards target states. Our experiments demonstrate the effectiveness of our model for learning optimal control of unknown dynamical systems. Codes available at https://github.com/chichengmessi/neural_ode_control/tree/main

Cheng Chi, Fangyun Wei, Han Hu

Existing object detection frameworks are usually built on a single format of object/part representation, i.e., anchor/proposal rectangle boxes in RetinaNet and Faster R-CNN, center points in FCOS and RepPoints, and corner points in CornerNet. While these different representations usually drive the frameworks to perform well in different aspects, e.g., better classification or finer localization, it is in general difficult to combine these representations in a single framework to make good use of each strength, due to the heterogeneous or non-grid feature extraction by different representations. This paper presents an attention-based decoder module similar as that in Transformer~\cite{vaswani2017attention} to bridge other representations into a typical object detector built on a single representation format, in an end-to-end fashion. The other representations act as a set of \emph{key} instances to strengthen the main \emph{query} representation features in the vanilla detectors. Novel techniques are proposed towards efficient computation of the decoder module, including a \emph{key sampling} approach and a \emph{shared location embedding} approach. The proposed module is named \emph{bridging visual representations} (BVR). It can perform in-place and we demonstrate its broad effectiveness in bridging other representations into prevalent object detection frameworks, including RetinaNet, Faster R-CNN, FCOS and ATSS, where about $1.5\sim3.0$ AP improvements are achieved. In particular, we improve a state-of-the-art framework with a strong backbone by about $2.0$ AP, reaching $52.7$ AP on COCO test-dev. The resulting network is named RelationNet++. The code will be available at https://github.com/microsoft/RelationNet2.

Shifeng Zhang, Cheng Chi, Yongqiang Yao et al.

Object detection has been dominated by anchor-based detectors for several years. Recently, anchor-free detectors have become popular due to the proposal of FPN and Focal Loss. In this paper, we first point out that the essential difference between anchor-based and anchor-free detection is actually how to define positive and negative training samples, which leads to the performance gap between them. If they adopt the same definition of positive and negative samples during training, there is no obvious difference in the final performance, no matter regressing from a box or a point. This shows that how to select positive and negative training samples is important for current object detectors. Then, we propose an Adaptive Training Sample Selection (ATSS) to automatically select positive and negative samples according to statistical characteristics of object. It significantly improves the performance of anchor-based and anchor-free detectors and bridges the gap between them. Finally, we discuss the necessity of tiling multiple anchors per location on the image to detect objects. Extensive experiments conducted on MS COCO support our aforementioned analysis and conclusions. With the newly introduced ATSS, we improve state-of-the-art detectors by a large margin to $50.7\%$ AP without introducing any overhead. The code is available at https://github.com/sfzhang15/ATSS

Sining Ang, Yuguang Yang, Chenxu Dang et al.

Vision-Language-Action (VLA) driving augments end-to-end (E2E) planning with language-enabled backbones, yet it remains unclear what changes beyond the usual accuracy--cost trade-off. We revisit this question with 3--RQ analysis in RecogDrive by instantiating the system with a full VLM and vision-only backbones, all under an identical diffusion Transformer planner. RQ1: At the backbone level, the VLM can introduce additional subspaces upon the vision-only backbones. RQ2: This unique subspace leads to a different behavioral in some long-tail scenario: the VLM tends to be more aggressive whereas ViT is more conservative, and each decisively wins on about 2--3% of test scenarios; With an oracle that selects, per scenario, the better trajectory between the VLM and ViT branches, we obtain an upper bound of 93.58 PDMS. RQ3: To fully harness this observation, we propose HybridDriveVLA, which runs both ViT and VLM branches and selects between their endpoint trajectories using a learned scorer, improving PDMS to 92.10. Finally, DualDriveVLA implements a practical fast--slow policy: it runs ViT by default and invokes the VLM only when the scorer's confidence falls below a threshold; calling the VLM on 15% of scenarios achieves 91.00 PDMS while improving throughput by 3.2x. Code will be released.

Enshen Zhou, Jingkun An, Cheng Chi et al.

Spatial referring is a fundamental capability of embodied robots to interact with the 3D physical world. However, even with the powerful pretrained vision language models (VLMs), recent approaches are still not qualified to accurately understand the complex 3D scenes and dynamically reason about the instruction-indicated locations for interaction. To this end, we propose RoboRefer, a 3D-aware VLM that can first achieve precise spatial understanding by integrating a disentangled but dedicated depth encoder via supervised fine-tuning (SFT). Moreover, RoboRefer advances generalized multi-step spatial reasoning via reinforcement fine-tuning (RFT), with metric-sensitive process reward functions tailored for spatial referring tasks. To support SFT and RFT training, we introduce RefSpatial, a large-scale dataset of 20M QA pairs (2x prior), covering 31 spatial relations (vs. 15 prior) and supporting complex reasoning processes (up to 5 steps). In addition, we introduce RefSpatial-Bench, a challenging benchmark filling the gap in evaluating spatial referring with multi-step reasoning. Experiments show that SFT-trained RoboRefer achieves state-of-the-art spatial understanding, with an average success rate of 89.6%. RFT-trained RoboRefer further outperforms all other baselines by a large margin, even surpassing Gemini-2.5-Pro by 17.4% in average accuracy on RefSpatial-Bench. Notably, RoboRefer can be integrated with various control policies to execute long-horizon, dynamic tasks across diverse robots (e,g., UR5, G1 humanoid) in cluttered real-world scenes. Please see the project page at https://zhoues.github.io/RoboRefer.

Enshen Zhou, Qi Su, Cheng Chi et al.

Automatic detection and prevention of open-set failures are crucial in closed-loop robotic systems. Recent studies often struggle to simultaneously identify unexpected failures reactively after they occur and prevent foreseeable ones proactively. To this end, we propose Code-as-Monitor (CaM), a novel paradigm leveraging the vision-language model (VLM) for both open-set reactive and proactive failure detection. The core of our method is to formulate both tasks as a unified set of spatio-temporal constraint satisfaction problems and use VLM-generated code to evaluate them for real-time monitoring. To enhance the accuracy and efficiency of monitoring, we further introduce constraint elements that abstract constraint-related entities or their parts into compact geometric elements. This approach offers greater generality, simplifies tracking, and facilitates constraint-aware visual programming by leveraging these elements as visual prompts. Experiments show that CaM achieves a 28.7% higher success rate and reduces execution time by 31.8% under severe disturbances compared to baselines across three simulators and a real-world setting. Moreover, CaM can be integrated with open-loop control policies to form closed-loop systems, enabling long-horizon tasks in cluttered scenes with dynamic environments.

Yi Han, Cheng Chi, Enshen Zhou et al.

Vision-Language Models (VLMs) have shown remarkable capabilities in spatial reasoning, yet they remain fundamentally limited to qualitative precision and lack the computational precision required for real-world robotics. Current approaches fail to leverage metric cues from depth sensors and camera calibration, instead reducing geometric problems to pattern recognition tasks that cannot deliver the centimeter-level accuracy essential for robotic manipulation. We present TIGeR (Tool-Integrated Geometric Reasoning), a novel framework that transforms VLMs from perceptual estimators to geometric computers by enabling them to generate and execute precise geometric computations through external tools. Rather than attempting to internalize complex geometric operations within neural networks, TIGeR empowers models to recognize geometric reasoning requirements, synthesize appropriate computational code, and invoke specialized libraries for exact calculations. To support this paradigm, we introduce TIGeR-300K, a comprehensive tool-invocation-oriented dataset covering point transformations, pose estimation, and spatial compatibility verification, complete with tool invocation sequences and intermediate computations. Through a two-stage training pipeline combining supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT) with our proposed hierarchical reward design, TIGeR achieves SOTA performance on geometric reasoning benchmarks while demonstrating centimeter-level precision in real-world robotic manipulation tasks.

Yulin Luo, Chun-Kai Fan, Menghang Dong et al.

Building robots that can perceive, reason, and act in dynamic, unstructured environments remains a core challenge. Recent embodied systems often adopt a dual-system paradigm, where System 2 handles high-level reasoning while System 1 executes low-level control. In this work, we refer to System 2 as the embodied brain, emphasizing its role as the cognitive core for reasoning and decision-making in manipulation tasks. Given this role, systematic evaluation of the embodied brain is essential. Yet existing benchmarks emphasize execution success, or when targeting high-level reasoning, suffer from incomplete dimensions and limited task realism, offering only a partial picture of cognitive capability. To bridge this gap, we introduce RoboBench, a benchmark that systematically evaluates multimodal large language models (MLLMs) as embodied brains. Motivated by the critical roles across the full manipulation pipeline, RoboBench defines five dimensions-instruction comprehension, perception reasoning, generalized planning, affordance prediction, and failure analysis-spanning 14 capabilities, 25 tasks, and 6092 QA pairs. To ensure realism, we curate datasets across diverse embodiments, attribute-rich objects, and multi-view scenes, drawing from large-scale real robotic data. For planning, RoboBench introduces an evaluation framework, MLLM-as-world-simulator. It evaluate embodied feasibility by simulating whether predicted plans can achieve critical object-state changes. Experiments on 14 MLLMs reveal fundamental limitations: difficulties with implicit instruction comprehension, spatiotemporal reasoning, cross-scenario planning, fine-grained affordance understanding, and execution failure diagnosis. RoboBench provides a comprehensive scaffold to quantify high-level cognition, and guide the development of next-generation embodied MLLMs. The project page is in https://robo-bench.github.io.

Zhe Li, Cheng Chi, Yangyang Wei et al.

Natural language offers a natural interface for humanoid robots, but existing language-guided humanoid locomotion pipelines remain cumbersome and untrustworthy. They typically decode human motion, retarget it to robot morphology, and then track it with a physics-based controller. However, this multi-stage process is prone to cumulative errors, introduces high latency, and yields weak coupling between semantics and control. These limitations call for a more direct pathway from language to action, one that eliminates fragile intermediate stages. Therefore, we present RoboGhost, a retargeting-free framework that directly conditions humanoid policies on language-grounded motion latents. By bypassing explicit motion decoding and retargeting, RoboGhost enables a diffusion-based policy to denoise executable actions directly from noise, preserving semantic intent and supporting fast, reactive control. A hybrid causal transformer-diffusion motion generator further ensures long-horizon consistency while maintaining stability and diversity, yielding rich latent representations for precise humanoid behavior. Extensive experiments demonstrate that RoboGhost substantially reduces deployment latency, improves success rates and tracking precision, and produces smooth, semantically aligned locomotion on real humanoids. Beyond text, the framework naturally extends to other modalities such as images, audio, and music, providing a universal foundation for vision-language-action humanoid systems.

Gangwei Xu, Haotong Lin, Hongcheng Luo et al.

This paper presents Pixel-Perfect Depth, a monocular depth estimation model based on pixel-space diffusion generation that produces high-quality, flying-pixel-free point clouds from estimated depth maps. Current generative depth estimation models fine-tune Stable Diffusion and achieve impressive performance. However, they require a VAE to compress depth maps into latent space, which inevitably introduces \textit{flying pixels} at edges and details. Our model addresses this challenge by directly performing diffusion generation in the pixel space, avoiding VAE-induced artifacts. To overcome the high complexity associated with pixel-space generation, we introduce two novel designs: 1) Semantics-Prompted Diffusion Transformers (SP-DiT), which incorporate semantic representations from vision foundation models into DiT to prompt the diffusion process, thereby preserving global semantic consistency while enhancing fine-grained visual details; and 2) Cascade DiT Design that progressively increases the number of tokens to further enhance efficiency and accuracy. Our model achieves the best performance among all published generative models across five benchmarks, and significantly outperforms all other models in edge-aware point cloud evaluation.

Haoran Zhang, Shuanghao Bai, Wanqi Zhou et al.

Robotic grasping is one of the most fundamental tasks in robotic manipulation, and grasp detection/generation has long been the subject of extensive research. Recently, language-driven grasp generation has emerged as a promising direction due to its practical interaction capabilities. However, most existing approaches either lack sufficient reasoning and generalization capabilities or depend on complex modular pipelines. Moreover, current grasp foundation models tend to overemphasize dialog and object semantics, resulting in inferior performance and restriction to single-object grasping. To maintain strong reasoning ability and generalization in cluttered environments, we propose VCoT-Grasp, an end-to-end grasp foundation model that incorporates visual chain-of-thought reasoning to enhance visual understanding for grasp generation. VCoT-Grasp adopts a multi-turn processing paradigm that dynamically focuses on visual inputs while providing interpretable reasoning traces. For training, we refine and introduce a large-scale dataset, VCoT-GraspSet, comprising 167K synthetic images with over 1.36M grasps, as well as 400+ real-world images with more than 1.2K grasps, annotated with intermediate bounding boxes. Extensive experiments on both VCoT-GraspSet and real robot demonstrate that our method significantly improves grasp success rates and generalizes effectively to unseen objects, backgrounds, and distractors. More details can be found at https://zhanghr2001.github.io/VCoT-Grasp.github.io.

Amine M. Aboussalah, Xuanze Li, Cheng Chi et al.

In the realm of option pricing, existing models are typically classified into principle-driven methods, such as solving partial differential equations (PDEs) that pricing function satisfies, and data-driven approaches, such as machine learning (ML) techniques that parameterize the pricing function directly. While principle-driven models offer a rigorous theoretical framework, they often rely on unrealistic assumptions, such as asset processes adhering to fixed stochastic differential equations (SDEs). Moreover, they can become computationally intensive, particularly in high-dimensional settings when analytical solutions are not available and thus numerical solutions are needed. In contrast, data-driven models excel in capturing market data trends, but they often lack alignment with core financial principles, raising concerns about interpretability and predictive accuracy, especially when dealing with limited or biased datasets. This work proposes a hybrid approach to address these limitations by integrating the strengths of both principled and data-driven methodologies. Our framework combines the theoretical rigor and interpretability of PDE-based models with the adaptability of machine learning techniques, yielding a more versatile methodology for pricing a broad spectrum of options. We validate our approach across different volatility modeling approaches-both with constant volatility (Black-Scholes) and stochastic volatility (Heston), demonstrating that our proposed framework, Finance-Informed Neural Network (FINN), not only enhances predictive accuracy but also maintains adherence to core financial principles. FINN presents a promising tool for practitioners, offering robust performance across a variety of market conditions.

Enshen Zhou, Cheng Chi, Yibo Li et al.

Spatial tracing, as a fundamental embodied interaction ability for robots, is inherently challenging as it requires multi-step metric-grounded reasoning compounded with complex spatial referring and real-world metric measurement. However, existing methods struggle with this compositional task. To this end, we propose RoboTracer, a 3D-aware VLM that first achieves both 3D spatial referring and measuring via a universal spatial encoder and a regression-supervised decoder to enhance scale awareness during supervised fine-tuning (SFT). Moreover, RoboTracer advances multi-step metric-grounded reasoning via reinforcement fine-tuning (RFT) with metric-sensitive process rewards, supervising key intermediate perceptual cues to accurately generate spatial traces. To support SFT and RFT training, we introduce TraceSpatial, a large-scale dataset of 30M QA pairs, spanning outdoor/indoor/tabletop scenes and supporting complex reasoning processes (up to 9 steps). We further present TraceSpatial-Bench, a challenging benchmark filling the gap to evaluate spatial tracing. Experimental results show that RoboTracer surpasses baselines in spatial understanding, measuring, and referring, with an average success rate of 79.1%, and also achieves SOTA performance on TraceSpatial-Bench by a large margin, exceeding Gemini-2.5-Pro by 36% accuracy. Notably, RoboTracer can be integrated with various control policies to execute long-horizon, dynamic tasks across diverse robots (UR5, G1 humanoid) in cluttered real-world scenes. See the project page at https://zhoues.github.io/RoboTracer.

Chenyu Lin, Cheng Chi, Jinlin Wu et al.

When confronted with complex problems, we tend to think slowly; conversely, for simple questions, we think quickly. Such a two-system thinking mechanism allows us to efficiently allocate cognitive resources, enabling quick decision-making for straightforward issues while reserving deeper analytical thinking for more intricate challenges. However, existing reasoning-oriented visual language models (VLMs), whether trained with explicit chain-of-thought annotations or rule-based RL rewards, mainly pursue lengthy, detailed reasoning chains, which often lead to excessive computational costs. In this work, we propose a simple RL approach, which enables VLMs to automatically switch between fast and slow thinking modes depending on task difficulty. The approach consists of two stages: in the first stage, we label data as either requiring fast thinking or slow thinking based on the model output length, which is inspired by the observation that pre-trained VLMs typically produce answers of varying lengths for different types of questions; in the second stage, we train the model using GRPO along with the thinking mode labels to develop dual-mode thinking. Despite its simplicity, our model, named DualMindVLM, significantly outperforms the base model and achieves performance on par with state-of-the-art visual reasoning models, while maintaining exceptionally high token efficiency.

Cheng Peng, Zhenzhe Zhang, Cheng Chi et al.

Navigating to a specified object in an unknown environment is a fundamental yet challenging capability of embodied intelligence. However, current methods struggle to balance decision frequency with intelligence, resulting in decisions lacking foresight or discontinuous actions. In this work, we propose PIGEON: Point of Interest Guided Exploration for Object Navigation with VLM, maintaining a lightweight and semantically aligned snapshot memory during exploration as semantic input for the exploration strategy. We use a large Visual-Language Model (VLM), named PIGEON-VL, to select Points of Interest (PoI) formed during exploration and then employ a lower-level planner for action output, increasing the decision frequency. Additionally, this PoI-based decision-making enables the generation of Reinforcement Learning with Verifiable Reward (RLVR) data suitable for simulators. Experiments on classic object navigation benchmarks demonstrate that our zero-shot transfer method achieves state-of-the-art performance, while RLVR further enhances the model's semantic guidance capabilities, enabling deep reasoning during real-time navigation.

Zeyi Liu, Cheng Chi, Eric Cousineau et al.

Audio signals provide rich information for the robot interaction and object properties through contact. This information can surprisingly ease the learning of contact-rich robot manipulation skills, especially when the visual information alone is ambiguous or incomplete. However, the usage of audio data in robot manipulation has been constrained to teleoperated demonstrations collected by either attaching a microphone to the robot or object, which significantly limits its usage in robot learning pipelines. In this work, we introduce ManiWAV: an 'ear-in-hand' data collection device to collect in-the-wild human demonstrations with synchronous audio and visual feedback, and a corresponding policy interface to learn robot manipulation policy directly from the demonstrations. We demonstrate the capabilities of our system through four contact-rich manipulation tasks that require either passively sensing the contact events and modes, or actively sensing the object surface materials and states. In addition, we show that our system can generalize to unseen in-the-wild environments by learning from diverse in-the-wild human demonstrations.

Cheng Chi, Shuran Song

This paper tackles the task of category-level pose estimation for garments. With a near infinite degree of freedom, a garment's full configuration (i.e., poses) is often described by the per-vertex 3D locations of its entire 3D surface. However, garments are also commonly subject to extreme cases of self-occlusion, especially when folded or crumpled, making it challenging to perceive their full 3D surface. To address these challenges, we propose GarmentNets, where the key idea is to formulate the deformable object pose estimation problem as a shape completion task in the canonical space. This canonical space is defined across garments instances within a category, therefore, specifies the shared category-level pose. By mapping the observed partial surface to the canonical space and completing it in this space, the output representation describes the garment's full configuration using a complete 3D mesh with the per-vertex canonical coordinate label. To properly handle the thin 3D structure presented on garments, we proposed a novel 3D shape representation using the generalized winding number field. Experiments demonstrate that GarmentNets is able to generalize to unseen garment instances and achieve significantly better performance compared to alternative approaches.

Cheng Chi, Dmitry Berenson

Estimating the state of a deformable object is crucial for robotic manipulation, yet accurate tracking is challenging when the object is partially-occluded. To address this problem, we propose an occlusion-robust RGBD sequence tracking framework based on Coherent Point Drift (CPD). To mitigate the effects of occlusion, our method 1) Uses a combination of locally linear embedding and constrained optimization to regularize the output of CPD, thus enforcing topological consistency when occlusions create disconnected pieces of the object; 2) Reasons about the free-space visible by an RGBD sensor to better estimate the prior on point location and to detect tracking failures during occlusion; and 3) Uses shape descriptors to find the most relevant previous state of the object to use for tracking after a severe occlusion. Our method does not rely on physics simulation or a physical model of the object, which can be difficult to obtain in unstructured environments. Despite having no physical model, our experiments demonstrate that our method achieves improved accuracy in the presence of occlusion as compared to a physics-based CPD method while maintaining adequate run-time.

Xiaobo Wang, Shuo Wang, Cheng Chi et al.

In face recognition, designing margin-based (e.g., angular, additive, additive angular margins) softmax loss functions plays an important role in learning discriminative features. However, these hand-crafted heuristic methods are sub-optimal because they require much effort to explore the large design space. Recently, an AutoML for loss function search method AM-LFS has been derived, which leverages reinforcement learning to search loss functions during the training process. But its search space is complex and unstable that hindering its superiority. In this paper, we first analyze that the key to enhance the feature discrimination is actually \textbf{how to reduce the softmax probability}. We then design a unified formulation for the current margin-based softmax losses. Accordingly, we define a novel search space and develop a reward-guided search method to automatically obtain the best candidate. Experimental results on a variety of face recognition benchmarks have demonstrated the effectiveness of our method over the state-of-the-art alternatives.

Cheng Chi, Shifeng Zhang, Junliang Xing et al.

Head and human detection have been rapidly improved with the development of deep convolutional neural networks. However, these two tasks are often studied separately without considering their inherent correlation, leading to that 1) head detection is often trapped in more false positives, and 2) the performance of human detector frequently drops dramatically in crowd scenes. To handle these two issues, we present a novel joint head and human detection network, namely JointDet, which effectively detects head and human body simultaneously. Moreover, we design a head-body relationship discriminating module to perform relational learning between heads and human bodies, and leverage this learned relationship to regain the suppressed human detections and reduce head false positives. To verify the effectiveness of the proposed method, we annotate head bounding boxes of the CityPersons and Caltech-USA datasets, and conduct extensive experiments on the CrowdHuman, CityPersons and Caltech-USA datasets. As a consequence, the proposed JointDet detector achieves state-of-the-art performance on these three benchmarks. To facilitate further studies on the head and human detection problem, all new annotations, source codes and trained models will be public.

Cheng Chi, Shifeng Zhang, Junliang Xing et al.

Pedestrian detection in crowded scenes is a challenging problem, because occlusion happens frequently among different pedestrians. In this paper, we propose an effective and efficient detection network to hunt pedestrians in crowd scenes. The proposed method, namely PedHunter, introduces strong occlusion handling ability to existing region-based detection networks without bringing extra computations in the inference stage. Specifically, we design a mask-guided module to leverage the head information to enhance the feature representation learning of the backbone network. Moreover, we develop a strict classification criterion by improving the quality of positive samples during training to eliminate common false positives of pedestrian detection in crowded scenes. Besides, we present an occlusion-simulated data augmentation to enrich the pattern and quantity of occlusion samples to improve the occlusion robustness. As a consequent, we achieve state-of-the-art results on three pedestrian detection datasets including CityPersons, Caltech-USA and CrowdHuman. To facilitate further studies on the occluded pedestrian detection in surveillance scenes, we release a new pedestrian dataset, called SUR-PED, with a total of over 162k high-quality manually labeled instances in 10k images. The proposed dataset, source codes and trained models will be released.

Shifeng Zhang, Cheng Chi, Zhen Lei et al.

Face detection has achieved significant progress in recent years. However, high performance face detection still remains a very challenging problem, especially when there exists many tiny faces. In this paper, we present a single-shot refinement face detector namely RefineFace to achieve high performance. Specifically, it consists of five modules: Selective Two-step Regression (STR), Selective Two-step Classification (STC), Scale-aware Margin Loss (SML), Feature Supervision Module (FSM) and Receptive Field Enhancement (RFE). To enhance the regression ability for high location accuracy, STR coarsely adjusts locations and sizes of anchors from high level detection layers to provide better initialization for subsequent regressor. To improve the classification ability for high recall efficiency, STC first filters out most simple negatives from low level detection layers to reduce search space for subsequent classifier, then SML is applied to better distinguish faces from background at various scales and FSM is introduced to let the backbone learn more discriminative features for classification. Besides, RFE is presented to provide more diverse receptive field to better capture faces in some extreme poses. Extensive experiments conducted on WIDER FACE, AFW, PASCAL Face, FDDB, MAFA demonstrate that our method achieves state-of-the-art results and runs at $37.3$ FPS with ResNet-18 for VGA-resolution images.

Chen Change Loy, Dahua Lin, Wanli Ouyang et al.

This paper presents a review of the 2018 WIDER Challenge on Face and Pedestrian. The challenge focuses on the problem of precise localization of human faces and bodies, and accurate association of identities. It comprises of three tracks: (i) WIDER Face which aims at soliciting new approaches to advance the state-of-the-art in face detection, (ii) WIDER Pedestrian which aims to find effective and efficient approaches to address the problem of pedestrian detection in unconstrained environments, and (iii) WIDER Person Search which presents an exciting challenge of searching persons across 192 movies. In total, 73 teams made valid submissions to the challenge tracks. We summarize the winning solutions for all three tracks. and present discussions on open problems and potential research directions in these topics.

Cheng Chi, Shifeng Zhang, Junliang Xing et al.

High performance face detection remains a very challenging problem, especially when there exists many tiny faces. This paper presents a novel single-shot face detector, named Selective Refinement Network (SRN), which introduces novel two-step classification and regression operations selectively into an anchor-based face detector to reduce false positives and improve location accuracy simultaneously. In particular, the SRN consists of two modules: the Selective Two-step Classification (STC) module and the Selective Two-step Regression (STR) module. The STC aims to filter out most simple negative anchors from low level detection layers to reduce the search space for the subsequent classifier, while the STR is designed to coarsely adjust the locations and sizes of anchors from high level detection layers to provide better initialization for the subsequent regressor. Moreover, we design a Receptive Field Enhancement (RFE) block to provide more diverse receptive field, which helps to better capture faces in some extreme poses. As a consequence, the proposed SRN detector achieves state-of-the-art performance on all the widely used face detection benchmarks, including AFW, PASCAL face, FDDB, and WIDER FACE datasets. Codes will be released to facilitate further studies on the face detection problem.